Would there be a way to “prove” that a given prompt/completion with Temperature=0 was generated with OpenAI?
Like, say you pay a few extra tokens, and OpenAI will write the prompt hash and the hash of the result to some kind of public ledger? So, given the prompt and completion, a client could re-hash them and compare them to the public version for confirmation?
Actually if you have Temp at zero, and you have the exact prompt then you should get a duplicate of the prompt/code.
As far as a Hash, technically AI is a Hash that creates. At the moment, of course.
To actually prove it, I think would be a difficult technical challenge without placing trust in at least OpenAI. They could easily create something that records the hash of a prompt/completion on a blockchain, but what would be the purpose?
You can find reddit post made by people doing this reverse engineering. Transformer models are not good at keeping secrets
Read up and follow rules. If you don’t understand read again. Don’t be afraid to reach out! Waste, fraud and abuse can happen on this as well as any platform but just as smokey the bear says with forest fires only you can prevent it from happening.